Soft Exosuit for Hip Assistance

Alan T. Asbecka,b, Kai Schmidta,b, Conor J. Walsha,b,∗

aBiodesign Lab, School of Engineering and Applied Sciences, Harvard University, Cambridge, MA, USAbThe Wyss Institute for Biologically-Inspired Engineering, Cambridge, MA, USA

Abstract

Exoskeletons comprised of rigid load-bearing structures have been developed for many years, but a newparadigm is to create “exosuits” that apply tensile forces to the body using textiles and utilize the bodysskeletal structure to support compressive forces. Exosuits are intended to augment the musculature byproviding small to moderate levels of assistance at appropriate times in the walking cycle. They have anumber of substantial benefits: with their fabric construction, exosuits eliminate problems of needing toalign a rigid frame precisely with the biological joints and their inertia can be extremely low. In this paper,we present a fully portable hip-assistance exosuit that uses a backpack frame to attach to the torso, ontowhich is mounted a spooled-webbing actuator that connects to the back of the users thigh. The actuators,powered by a geared brushless motor connected to a spool via a timing belt, wind up seat-belt webbing ontothe spool so that a large travel is possible with a simple, compact mechanism. Designed to be worn overthe clothing, the webbing creates a large moment arm around the hip that provides torques in the sagittalplane of up to 30% of the nominal biological torques for level-ground walking. Due to its soft design, thesystem does not restrict the motion of the hip in the ab- and adduction directions or rotation about the legaxis. Here we present the design of the system along with some initial measurements of the system in useduring walking on level ground at 1.25 m/s, where it creates a force of up to 150 N on the thigh, equivalentto a torque of 20.5 Nm to assist hip extension.

Keywords: exosuit, wearable robotics, walking assistance

1. Introduction

A large number of lower limb exoskeletons havebeen developed over the years, usually with the pur-pose of assisting or augmenting human walking. Forindividuals needing to carry heavy loads such assoldiers or recreational backpackers, the promise ofa mechanical assist is welcoming, potentially reduc-ing muscle fatigue, metabolic expenditure, or injuryrates. For individuals needing assistance with walk-ing such as the elderly or paraplegic, or patientsrequiring gait rehabilitation, such devices could po-tentially restore walking function.

Many previous exoskeletons have comprised ofrigid load-bearing structures, designed to trans-mit forces to the ground while tracking or apply-

∗Corresponding author.Email addresses: aasbeck@gmail.com (Alan T.

Asbeck), kschmidt@seas.harvard.edu (Kai Schmidt),walsh@seas.harvard.edu (Conor J. Walsh)

ing torques to the wearer’s joints at points alongthe structure. Some of these use the exoskele-ton to support the weight of a hiker’s backpack[1, 2, 3, 4], while others support a paraplegic’s body-weight [5, 6, 7, 8, 9, 10].

An alternative approach that has been appliedto assist walking is to apply torques to the user’sjoints in parallel with the musculature, but with-out transmitting a load to ground. In particu-lar, devices have been constructed to augment thestrength of the hip and knee in healthy individu-als [11, 12]. Many other such devices are poweredorthoses [13, 14, 15] that are designed to supportdisabled individuals by providing assistance at boththe hip and knee to compensate for reduced musclestrength. These systems are also rigid, and may re-quire an ankle-foot orthosis to be used with themto provide stability for the ankle joint and preventmigration of the rigid components that can oftenhave significant weight. Two previous devices for

Preprint submitted to Robotics and Autonomous Systems November 3, 2014

assisting the hip in flexion used pneumatic actua-tors [16, 17, 18] along with rigid plastic orthoses tosecure to the leg and pelvis such that a pin jointwas approximately colocated with the hip. Severalother devices have been made to provide assistanceto the ankle alone, both for plantarflexion and dor-siflexion [19, 20]. These units offer the potential toincrease power generation capability at the ankleto support forward propulsion in addition to ensur-ing sufficient toe clearance from the ground duringswing.

A more recent rigid exoskeleton is the HondaWalking Assist Device with Stride ManagementSystem [21, 22], which actuates the hip joint inboth flexion and extension. This device consistsof a hybrid soft-rigid waist belt and struts that ex-tend partially down the thigh, where they attachwith narrow straps. It is powered by a motor ap-proximately colocated with the wearers hip jointin the sagittal plane and includes a passive jointjust above the motor to permit small amount ofhip ab-/adduction. It functions differently from thepreviously-discussed examples, in that it only ap-plies small forces and serves primarily to regulatethe cadence of the wearer. Humans are known towalk most efficiently at a certain cadence (steps perminute) for a given forward velocity, and the Hondasystem applies small torques to the hips to enticethe user into walking at that cadence.

As an alternative to rigid exoskeletons, a newparadigm is to create “exosuits” that are comprisedof fabrics and utilize the body’s skeletal structureto support compressive forces. Exosuits are usedto augment the musculature by providing smallamounts of assistance at crucial times in the walk-ing cycle, as opposed to being able to apply largetorques or support significant weight. These sys-tems have a number of substantial benefits: withtheir fabric construction, exosuits eliminate prob-lems of needing to align a rigid frame precisely withthe biological joints and their inertia can be ex-tremely low. These two features virtually eliminateresistance to motion, thus permitting close to nat-ural kinematics. Furthermore, exosuits can be lightand sleek, permitting them to be worn constantly;thus enabling rehabilitation or strength augmenta-tion to occur for longer periods of time throughoutthe day. However, while these systems do offer po-tential benefits over rigid exoskeletons, there arelimitations related to controllability and maximumapplied force and thus much research remains tocharacterize them and understand their capabili-

ties.

While the concept of an exosuit is relatively new,a number of designs have already been developed.Wehner et al. [23] created a pneumatically-poweredsystem that assists ankle plantarflexion, and usesa garment of cloth and webbing to secure to thelegs, waist, and shoulders. Asbeck et al. [24] de-veloped a Bowden-cable-driven biologically inspiredexosuit that assists both ankle plantarflexion andhip flexion with a single actuator for each leg. Thisis achieved with a multi-articular exosuit architec-ture that transmits forces from the waist and thighthrough the knee and to the back of the calf, whereit connects to the heel via the Bowden cable. Kawa-mura et al. [25] made a pneumatically-powered exo-suit to assist hip flexion, with the device attachingto the wearer with a wide waist belt and a bracethat crosses the knee. In addition to assisting walk-ing, several other exosuits have been developed thatsupport the back and torso while the wearer is lift-ing heavy loads [26, 27, 28].

The system described in this paper is a new ex-ample of a soft exosuit that uses the body’s bonestructure to support compressive forces and doesnot provide kinematic restrictions to the joints. Thedesign assists hip extension, and is shown in figure1.

In this paper, we focus on the design of theexosuit. In the following sections, we presentan overview of the system, model how it appliestorques to the body, and compute the requirementsfor the actuators. We describe and perform anal-ysis for a new spooled-webbing actuator. We alsopresent preliminary measurements of the system inuse during walking on level ground, demonstratingits functionality.

2. System overview

Our design (figure 1) utilizes a backpack frameto attach to the torso, and uses geared motors toretract webbing ribbon connected to a thigh braceon each leg. The actuators wind up the webbingribbon onto a spool, which permits large (>25 cm)travel with a simple mechanism. The bottom offigure 1 shows how the device functions during thewalking cycle. The actuators retract the webbingjust before heelstrike, which in conjunction withthe leg’s motion creates forces pulling the thighback. The force peaks just after heelstrike, concur-rent with the wearer extending their hip to support

2

Figure 1: Top, exosuit to assist hip extension described inthis paper. Two actuator units are mounted on a backpackframe, and connect to cloth thigh braces with webbing. Bot-tom, overview of device operation. Starting at 90% in thegait cycle, which extends from one heelstrike to the next, theactuator units retract the webbing. This produces a force onthe thigh that increases until 6% in the gait cycle. The actu-ator units then spool out the webbing so the force decreases,reaching zero at 20% in the gait cycle. At other times in thegait cycle, the webbing is slack so that no force is applied tothe leg.

themselves on the newly-planted leg. The actua-tors then extend the webbing again, so that thereis no force in the webbing for the majority of thegait cycle.

This actuation scheme is unique in that the rib-bon is wound on top of itself, increasing the spooldiameter as the ribbon retracts. Mechanisms thatwind or unwind a thin ribbon from a spool withan actuator have been used in several applicationspreviously, including retracting seat belts, readingaudio and video magnetic tapes, mopping floors,and deploying drip system tubing [29, 30, 31, 32].For comparison, many devices wind up a thin ca-ble around a spool, but the cable typically travelsdown the length of the spool as it winds, insteadof solely winding on top of itself as it does in ouractuators. We use webbing instead of a thin cableso the device is comfortable if the wearer is seatedor if their leg is at large (> 90◦) angles in flexion.In these scenarios, a narrow cable would dig intothe body, while the 2” wide webbing will press inonly a small amount due to its much larger contactarea. The webbing further forms a wide attachmentto the thigh brace, helping distribute forces aroundthe thigh.

The webbing path between the backpack andthigh brace creates a large moment arm around thehip, providing torques in the sagittal plane of upto 30% of the nominal biological hip torques forlevel-ground walking. A key feature of this soft de-sign is that it does not restrict the motion of thehip in the ab- and adduction directions or in rota-tion about the leg axis. The webbing also permitsthe force to be measured inside the actuator unit,since it extends directly from it to the thigh attach-ment. Measuring forces internally permits the unitto be compact and robust, with no wiring or sen-sors outside of the actuation units. The system alsoincludes two footswitches that are used to estimatethe wearer’s cadence and phase through gait.

The prototype device is capable of applyingforces to the leg of up to 150 N at walking speedsup to 1.79 m/s (4 mph). The device utilizes thebackpack frame to transmit torques to the torso,pushing forward and downward on the lower backwith forces up to 188 N and pulling back on theshoulders with forces up to 50 N. The entire de-vice weighs 7.57 kg including batteries to last for2.8 hours of use. The bulk of the weight is on thetorso, with each thigh brace providing minimal in-ertia to the leg with a weight of 0.17 kg.

3

3. Calculations

In this section we present a simple model of thesystem and perform calculations to determine thenecessary velocity of the ribbon at the motor inorder for the actuators to achieve sufficient forcesnecessary to apply a scaled version of the nomi-nal hip moment during walking on level ground at1.43 m/s (3.2 mph). It should be noted that thesevalues depend on the compliance of the body andsuit in addition to the leg’s motion.

3.1. System Modeling

A diagram of the system is shown in figure 2.This shows the backpack frame worn on the user’storso and the webbing ribbon extending downwardto the thigh brace. Within the leg, a number ofsprings are shown, representing the compliance ofthe tissue between the surface of the leg and theunderlying bone. The muscle, fat, and other tis-sue in the leg will compress as they transfer forcesto the femur bone, which is the only rigid compo-nent in the leg. Additional springs are shown atthe waist and the shoulder, where the body alsohas some compliance. We model the total effectivestiffness of all of these locations as they interactwith the system together as the stiffness kbody. Thecompliance of the ribbon itself can be ignored sinceit is comparatively inextensible, with a stiffness of600,000 N/m for a 25 cm length (Young’s modulus= 2.41 GPa).

The top right of the diagram in figure 2 showsa geometric model of the wearer’s leg and how theexosuit attaches to it. The pivot point of the hip inthe sagittal plane is shown by a large black dot. Theleg extends downward from this point at an angleθhip with respect to vertical (shown in the figure aspositive), and the leg bone forms an angle ψ withrespect to the webbing ribbon. The pivot point ofthe hip is assumed to be horizontal from the bottomof the actuators where the webbing exits, and thedistance between these two points is denoted a. Thedistance between the hip pivot point and the pointat which the ribbon would intersect the femur bone,if it continued through the leg tissue, is denoted l.As the user walks, the perpendicular moment arm rbetween the ribbon and the hip joint will vary dueto θhip changing. The force on the webbing, F , andthe moment arm r, combine to form the momentexerted on the hip, Mhip.

The bottom right of the diagram shows the dis-tances related to the ribbon travel. In each case,

Figure 2: Left, diagram of person wearing the system, show-ing the forces exerted on the wearer by the device. Thewearers body compresses at the thigh, waist, and shoulder,the behavior of which is represented by the lumped stiffnesskbody . Right, model used to calculate the required motordisplacement in order to apply a specified moment Mhip atthe hip as the hip moves.

these distances are along the ribbon, which willchange orientation as the wearer walks. xm is theposition of the ribbon at the actuator, with positiveout of the actuator (downward and to the right inthe figure). xhip is the position of the ribbon atthe virtual point where it connects to the femur,also with positive downward and to the right, withxhip = 0 when the hip is furthest in extension. Inbetween, xs is the length of the simulated springrepresenting the compliance of the leg tissue kbody.While in reality the leg tissue is compressing, this isequivalent to an extension spring extending back-ward from the femur, and it is drawn as such in themodel for simplicity. For each of these positions, weignore constant offsets since we are only concernedwith the net travel of the tissue and actuator.

The length of ribbon between the actuator andfemur is defined as:

lribbon =√a2 + l2 − 2al cos (π/2 + θhip) (1)

The travel of the ribbon needed to follow the

4

hip motion (xhip) is a function of this distance.Subtracting

√a2 + l2 so that xhip(θhip = 0) = 0,

we get:

xhip = lribbon −√a2 + l2 (2)

To calculate the external moment Mhip that thehip exosuit is capable of applying to the person, itis necessary to determine the angle ψ between thethigh and the webbing.

Mhip = rF (3)

r = l sin(ψ) (4)

ψ = arccos

(l2 + l2ribbon − a2

2 l lribbon

)(5)

⇒ F =Mhip

l sin (ψ)(6)

Beyond modeling the geometric changes as thehip moves during walking, the stiffness of the bodymust be taken into account. Intuitively, in addi-tion to tracking the hip’s motion, to apply forcesthe motor must retract the ribbon additionally tocompress the thigh tissue.

Experiments discussed in section 3.2 found thatthe body stiffness can be approximated by a second-order equation:

F (xs) = p1x2s + p2xs (7)

For a force to be applied to the webbing, it mustdisplace the suit-human spring kbody:

F = p1x2s + p2xs ≡

Mhip

l sin (ψ)(8)

⇒ xs =−p22p1

+

√(p22p1

)2

+

(Mhip

p1l sin (ψ)

)(9)

Finally, the motor position xm can be calculatedfrom the sum of the hip motion and the spring mo-tion. According to the defined coordinate frame,the motor position xm that is required to track andapply forces to the hip can be calculated by:

xm = xhip − xs (10)

We use this modeled motor position to determinerequirements for the actuator units, and as a basisfor generating motor trajectories in practice.

Figure 3: Measured stiffness of the suit-human system, de-termined by retracting the webbing while holding the hipin a fixed vertical position and measuring the force at theactuator unit. Three cycles of a sine wave actuation profileare shown with thin black lines. Also shown is a quadraticline of best fit to the rising edge (thicker dashed line).

3.2. Measurements of Suit-Human Stiffness

To determine the actual value of the lumped stiff-ness kbody, a set of experiments were performed us-ing the assembled device. This and other human-subjects experiments in this paper were approvedby the Harvard Institutional Review Board. Forthe protocol, the wearer stood with the measuredleg forward with their feet 40 cm apart as shown inthe inset in figure 3, and the actuator was com-manded to retract the ribbon 16.5 cm and thenrelease it following a sinusoidal profile at 0.5 Hz.The ribbon displacement was determined experi-mentally so that a force of approximately 200 Nwould be applied to the wearer. This was repeatedthree times, and the results plotted in figure 3 withan offset removed where no force was applied to thebody.

The rising portion of the curve (correspondingto when the actuator was pulling) was fitted to aquadratic equation, also plotted in figure 3, result-ing in the line of best fit:

F (xs) = p1x2s + p2xs (11)

p1 = 8899.7, p2 = 99.546 (12)

where xs is in meters and F in Newtons.

5

3.3. SimulationsUsing the model of the system and the measured

body stiffness, we next compute the required actu-ator travel in order to apply a scaled version of thenominal hip moment to the body.

We begin with the nominal biological sagittal hipmoment for level-ground walking, duplicated from[33]. This is plotted in figure 4(top) with respectto the percentage through the gait cycle, which ex-tends from one heel strike to the next for a givenfoot. The biological moment begins slightly be-fore heel strike to decelerate the leg, and peaksjust after heel strike to prevent the leg from col-lapsing as it accepts the weight of the body. Wenext form an approximated hip moment which issmoothed relative to the nominal moment and islimited to positive (extensor) torques, also shownin figure 4(top). This curve was constructed byselecting seven points along the nominal momentprofile to match the starting point, peak, end point,and several relevant points in between, then inter-polating and smoothing with a low-pass filter. Itdiffers slightly from the nominal moment between90-100% in the gait cycle to avoid rapid changesin force which an actuator would have difficultyfollowing, and because a smooth moment profilewill feel more comfortable to the wearer. Sincethe exosuit is designed to apply a partial assistancetorque to the wearer, we then scale the smoothedapproximation by 31% for use in subsequent calcu-lations. This scaling corresponds to a peak forceof 145 N at the leg for a 76 kg individual. Whenthe smoothed approximation is scaled by 31%, it isclose in magnitude to the unscaled biological mo-ment between 90% and 100% in the gait cycle. Weuse this scaled approximation as Mhip in our model.Using this along with the nominal biological hip an-gle (not shown), also from [33], and plugging theminto equations 2-8, we generate positions xhip, xs,and xm, with offsets removed so that xhip reaches0 when the hip is furthest back and xs is zerowhen no force is applied. These are shown in figure4(middle). In these calculations, we use a = 0.32 m,l = 0.266 m, and we assume a gait period of 1.15seconds which was empirically measured with a sub-ject walking on a treadmill at 1.25 m/s (2.8 mph).

In figure 4(middle), the hip position xhip startsat a large positive value at heel strike (the leg is for-ward, in flexion) and decreases until around halfwaythrough the gait cycle when the leg is furthest back,in extension, and then increases in value again. Ifthe motor were to perfectly track the hip’s motion,

Figure 4: Top, biological hip moment profile vs. approxi-mated profile, which is used for the simulation after scalingby 31%. Hip extension moments are positive, which corre-sponds to the muscles pulling the leg toward the posterior.Middle, resulting position profiles of the hip (xhip), spring(xs), and motor (xm) during ground level walking. Hereflexion positions are positive, which corresponds to the legangled towards the front of the body or the motor spool-ing out the ribbon. The spring position is plotted with thespring extending being positive. Bottom, hip speed (vhip)and motor speed (vm) that is necessary both to track andto actuate the hip. All graphs are shown as a function ofpercentage through the gait cycle, which extends from oneheel strike to the next.

applying no force to the ribbon, then the motortrajectory would follow this profile as well. In-stead, the motor must retract the ribbon addition-ally (more negative values of xm in the plot) togenerate force across the body’s compliance. As in-dicated on the far left of the plot, the difference inheight between the hip position xhip and the motorposition xm is the amount the spring is extended(xs). In the center of the gait cycle, when the ap-plied moment is zero, the simulation shows the mo-tor tracking the hip (xm = xhip), as indicated inthe figure.

In addition to the solid lines in the figure, a pos-sible alternate path for the motor position is shownin a red dashed line. This line is above the hip po-sition, which corresponds to slack in the ribbon.Slack in the ribbon guarantees that the force is

6

zero during 35-85% in the gait cycle; if the actua-tors were tracking the leg imperfectly, antagonisticforces could be applied to the leg during that time.When operating the device in practice, we use amotor trajectory similar to this that creates slackin the ribbon.

In figure 4(bottom), the resulting velocities of theribbon at the hip (vhip) and motor (vm) are plotted.The motor velocity at around 90% in the gait cycleshows that a speed of 0.95 m/s is required to com-press the body’s compliance and apply the specifiedmoment to the hip. For comparison, if the body wasextremely stiff, the actuators would have to followa path very close to that of the hip itself, and thushave a velocity of vhip. To track the hip’s motion,this would require a speed of 0.67 m/s, which occursat around 70% in the gait cycle when no force is onthe ribbon. During the period when force is appliedto the ribbon, the maximum speed of the leg is only0.32 m/s, a value much slower than 0.95 m/s. Insummary, a major effect of the body’s complianceis to increase the required motor speed for applyingforces by more than a factor of two.

One interesting point is how to optimize the posi-tioning of the ribbon with respect to the hip joint.The simulation assumed relatively large distancesbetween the actuator and the hip joint (a = 0.32 m)and the distance down the leg (l = 0.266 m). It isinteresting to consider what would happen if thesedistances were varied. This analysis is illustratedin figure 5, which shows the peak force F and peakmotor velocity vm required to achieve the approxi-mated moment profile in figure 4 as distances a andl are varied.

According to the measured stiffness data, thebody’s compliance requires 12 cm of ribbon travelto achieve 145 N of force. This displacement wouldlikely be similar if the distances a and l werechanged, since the combined stiffness of the leg,waist, and shoulder is largely independent of thestrap’s positioning on the leg. For small values of aand l, in the bottom left corner of the plots, largerribbon forces are required because the moment armabout the hip has been reduced. The motor veloc-ities are also increased, because the larger forcesrequire additional displacement of the body’s com-pliance, and it must occur in the same amount oftime as before. The motor would then be muchfaster than needed in order to track the hip’s mo-tion, which would be slower by a factor of two dueto the smaller moment arm. Thus, smaller valuesof a and l result in a higher peak motor power and

Figure 5: Peak force F (top) and peak motor velocity vm(bottom) required to produce the approximated hip momentshown in figure 4 after it is scaled as distances a and l arevaried. As the distance between the ribbon and the hipjoint increases (a and l increase), the required peak forcegoes down. The required motor velocity decreases up to acertain point due to the motor being more able to compressthe body’s compliance. For large values of a and l, the peakmotor velocity increases due to the necessity of tracking theleg at 70% in the gait cycle. The values of a and l used inthe device are shown by an asterisk in each plot.

a mismatch in required motor speeds at differentpoints in the gait cycle.

For larger values of a and l, the force decreasesmonotonically as the radius between the ribbon andhip joint increases. The peak motor velocity alsodecreases as a and l increase, but only up to a point.In the top right corner of the plot, with large val-ues of a and l, higher motor velocities are againneeded, but now in order to track the hip’s motionat around 70% in the gait cycle. In this region,the motor power during the pull phase is reduced,but the motor may be required to move faster thandesired. The minimum required motor velocity cor-responds to a ribbon position which results in near-

7

Figure 6: Left, assembled actuation unit, and center, view of the inside. Right, diagram of the pulley system, which measuresthe ribbon force F indirectly by measuring Fr . 1 - spool; 2 - idler; 3 - feeder system; 4 - cantilever load cell (Phidgets 20 kgMicro Load Cell); 5 - timing pulley; 6 - Maxon EC 4-pole 30 motor plus 32 mm gearbox

equal speeds for compressing the body’s complianceand tracking the joint’s motion.

The values of a and l used in our device areclose to this point, and were prevented from be-ing any larger due to practical concerns. The thighbrace was made tall (20 cm) to distribute the forcebroadly over the wearer’s thigh, which limited themaximum possible value of l. Similarly, the valueof a was restricted to prevent the device from pro-truding too far from the wearer.

For the device we constructed, we chose a maxi-mum speed of 0.75 m/s to design towards, slightlyslower than the 0.95 m/s predicted by the simu-lation, in order that the peak force delivered tothe body might be slightly higher than would bepossible with a 0.95 m/s peak speed. This slowerspeed means that the position and force pulse willbe slightly more spread out as compared to the sim-ulated one. This speed is still capable of trackingthe hip’s motion, however, which only requires aspeed of 0.67 m/s.

4. Mechanical Design

Following our determination of the required rib-bon speed at the motor, we can design the actuatoraccordingly. We first present the overall design ofthe actuator unit, and then discuss the ribbon spooland the electrical design.

4.1. Overall Design

A rendering of the actuator unit is shown in figure6. The unit consists of a geared motor (6) connected

to the spool (1) that winds up the webbing ribbonvia a timing belt (5). After exiting the spool, theribbon passes over an idler pulley connected to twocantilever-style load cells (4) before exiting the unitthrough a feeder system (3) comprised of two ad-ditional idlers. In addition to guiding the ribbon,this system of idlers and load cells is used to detectthe force in the ribbon, as shown in the far right offigure 6. The ribbon passes over the instrumentedidler at an angle β on each side, so that when theribbon pulls with force F , a horizontal force Fr isinduced on the instrumented idler. By measuringthis force with a load cell on each side of the idler,the ribbon force can be computed via

F =Fr

2 sin(β)(13)

where Fr is the sum of the readings from thetwo load cells. The load cells achieve an accuratereading of the ribbon force because the friction inthe feeder system is low, and the ribbon does notcontact anything between the feeder and where itattaches to the leg.

4.2. Ribbon Spool

The actuator functions by winding up webbingonto a spool, depicted in figure 7(top). Thespool has a minimum usable radius rs,min, a maxi-mum radius rs,max, and the webbing has thicknesstwebbing. The webbing exiting the spool is at posi-tion xm and pulls with force F , as previously dis-cussed. The shape of the webbing in the spool is

8

an Archimedes’ spiral [34], with radius r a functionof the angle the webbing forms around the spool φ.In general, an Archimedes spiral has the equation

r(φ) = αφ (14)

In our situation, we have α = twebbing/(2π), thespool has a minimum radius rs,min, and we denotethe radius at which the webbing exits the spoolrspool, resulting in

rspool(φ) =twebbing

2πφ+ rs,min (15)

If the spool rotates at angular velocity ω = dφdt ,

the linear velocity of the webbing exiting the spoolis

v = rspoolω =twebbing

2πφφ

dt+ rspool,min

φ

dt(16)

and the arc length between rs,min and rspool is

s(φspool) =

twebbing4π

((φspool + c)

√1 + (φspool + c)2

+ ln

((φspool + c) +

√1 + (φspool + c)2

)− c√

1 + c2 + ln c+√

1 + c2))

(17)

where

c =2π

twebbingrs,min (18)

These equations are derived by taking the arclength of an Archimedes’ spiral from [34], substi-tuting in the appropriate variables for our applica-tion, and subtracting the equivalent arc length ofthe center of the spool (r < rs,min).

Since the spool will change radius as it winds upor feeds out the ribbon, the velocity of the ribbonand torque produced by the actuator will changeslightly over time. Figure 7(middle) examines howthe radius changes for a generic spool as a func-tion of several parameters. The x-axis of this plotshows the rotations of a sample spool, with a fully-wound spool at radius rs,max having N turns, andeach tick mark to the left of that corresponding toone fewer winding of ribbon around the spool. Onthe vertical axis, the radius of the spool is plotted

Figure 7: Top, diagram of the spool depicting the relevantvariables. Middle, plot of how the spool radius varies as afunction of the number of rotations of the spool for variousribbon thicknesses. Data are normalized to be fractions ofthe maximum spool radius rmax. Bottom, webbing speedand force as a function of spool radius in our actuators.

9

as a fraction of the maximum spool radius. Theplot shows results for several different thicknessesof ribbon: the top line shows a very thin ribbonof thickness 0.01rs,max while the bottom line showsa thick ribbon (relative to the spool maximum ra-dius) of 0.1rs,max. For the thinnest ribbon shown,the radius varies by only 5% over five rotations ofthe spool, while for the thickest ribbon shown itvaries by 50%. Since the velocity of the ribbon ex-iting the spool is proportional to the spool radiusfor a constant angular velocity, this graph can alsobe used directly to understand the variation in ve-locity with a spooled ribbon. For most systems, athin ribbon is desired so the velocity and torque willnot vary significantly (< 10%) during operation.

4.2.1. Parameters for our system

Our device uses seatbelt webbing for the rib-bon, which has a thickness of twebbing = 1.26 mm.We chose a spool radius with rs,min = 2.75 cmand rs,max = 3 cm. The minimum radius is largeenough to provide space inside the spool to securethe webbing to the spool, and so that the ribbonvelocity is relatively constant. The size also min-imizes the inertia of the spool. With these num-bers, twebbing = 0.042rs,max, which would be lo-cated on figure 7(middle) just above the twebbing =0.05rs,max line. The spool radius is large enoughthat one rotation of the spool corresponds to ap-proximately 18 cm of ribbon travel, which is morethan enough to actuate the hip during level-groundwalking. This means that the change in radiusduring walking (and hence velocity) is less than5% throughout the gait cycle. The minimum andmaximum values of our spool radius correspond to1.98 wraps around the spool and 35.8 cm of ribbontravel. This travel gives extra room for adjustingthe device to different sizes of individual as well aspermitting the actuators to feed out enough ribbonto permit the wearer to bring their knee to theirchest, which requires 25 cm of ribbon.

The gear reduction can be computed after thespool radius is determined. The encoder on themotor is limited to 12000 rpm = 1256.6 rad/sec.With the output spool at its minimum radius of2.75 cm, and a required ribbon velocity of 0.75 m/s,the output spool must have an angular velocity ofω = v/rspool = 27.27 rad/sec. This gives a requiredgear reduction of 46.07:1 between the motor andoutput spool. We choose a gearbox of 23:1 anduse a timing belt to give another 2:1 reduction toachieve 46:1 for the combination.

Figure 8: Electronics block diagram.

To compute the expected ribbon force at the out-put of the actuator, we estimate the efficiency of thedrive train components. The Maxon 32 mm 23:1gearbox has a quoted efficiency of 75%, the timingbelt is assumed to have an efficiency of 95%, and theother idler pulleys and friction with the feeder havean assumed efficiency of 90-95%. This combines tomake a total mechanical efficiency of 64.1-67.7%.With a motor torque of 0.14 Nm, this leads to atorque at the minimum spool radius of 4.128 Nmand a force on the ribbon of 150 N. The final rib-bon speed and force are plotted as a function ofspool radius in figure 7(bottom).

4.3. Electronics Overview

A block diagram of the electronics is shown infigure 8. An Aurora PC104 realtime computer run-ning Matlab xPC Target is the primary controller.The PC104 reads in the amplified load cell signalsand the footswitches, and sends an analog positioncontrol signal to an mbed LCP1768 microcontrollerwhich acts as a communications module. This inturn runs a position controller, and uses a CAN busto transmit desired motor currents to the RoboteqSBL1360 motor controllers. To complete the feed-back loop, the motor controllers transmit the motor

10

encoder readings back to the mbed communicationsmodule, which then converts them to an analog sig-nal and sends them back to the PC104 for logging.

5. Results

For preliminary walking experiments, we usedfootswitches to segment the gait, and a positioncontroller to move the ribbon spool through a tra-jectory at specified times in the gait cycle. Fol-lowing our calculations of xm in section 3.3, weform a position trajectory profile by using a tri-angular pulse with a similar amplitude as xm (12.3cm actual vs. 11.8 cm xm) and similar timing, butwith the motor acceleration- and velocity-limitedto 3500 rotations/sec2 and 12000 rpm, respectively.The resulting profile extends from 80-25% in thegait cycle. The controller maintains a buffer of thelast 5 footstrike times and computes the averagegait period. It assumes the wearer is walking at aconstant velocity, and uses the average gait periodto estimate when they are at 80% in the gait cy-cle. At that point, the motor executes the position-controlled motor trajectory, which creates a forcein the ribbon through the compliance of the body.This scheme depends on the wearer walking at aconstant cadence in order to have the motor’s pullline up well with the biological hip moment. Evenso, this simple control scheme is useful for deter-mining the behavior of the device.

Results from an individual wearing the systemwhile walking at 1.25 m/s (2.8 mph) are shown infigure 9. In this figure, we compare simulated datawith measured data to validate the model. For thesimulated data, we use the reference hip angle θhipfrom [33], lengths a and l measured on the subject,and the actual motor trajectory xm. These are usedwith the model developed in section 3 to calculatethe resulting webbing force and hip position. Inthe figure, the top graph shows the actual motortrajectory xm along with the simulated hip positionxhip and spring length xs. The bottom plot showsthe simulated force as compared to the measuredforce and its standard deviation.

As shown in the bottom graph, in practice thedevice achieves 9% less force (133 N vs. 146 N)despite 0.5 mm additional cable travel as comparedto the simulation. This difference is likely due toadditional torso lean which occurs during walking,especially when the force is applied at the shoulder.The measured force also drops off faster than thesimulated force; this is likely due to the fact that the

Figure 9: Results for one leg for a subject walking at 1.25 m/s(2.8 mph). Top, graph of motor position xm used in theexperiment, and simulated xhip and xs. Bottom, graph ofmeasured force in the ribbon with standard deviation as com-pared to simulated force using the model.

model did not include the hysteresis in the body-suit stiffness curve. In general, though, the modelfits the measured data well.

The resulting assistive pulses of force feel natu-ral due to their being aligned well with the nomi-nal biological torques. During the period of forceapplication, the leg is approximately at a constantangle of Ψ = 35◦, which also can be seen from theapproximately constant position of xhip in figure 9from 90-15% in the gait cycle. At this angle, thethigh brace pulls up along the leg as well as back-ward. This does result in some translation of thethigh brace vertically along the leg, but this motionis minimal due to the conical shape of the thighjust above the knee. If the thigh brace is initiallywrapped snugly around the thigh, there is no mo-tion between the skin and thigh brace, making itcomfortable to wear during actuation for the levelof force we propose to apply.

6. Conclusion

In conclusion, we have modeled and designed anexosuit that effectively applies moments to the hip

11

joint, contributing up to 30% of the nominal biolog-ical moment for walking. The design does not re-strict the hip’s motion with rigid linkages, but relieson the bone structure to support compressive forcesacross the joint. Substantial forces can be appliedto the leg through the soft interface, and the limit-ing factor is not the comfort of the system but theavailable motor power. The soft interface does havesubstantial compliance, which requires higher mo-tor speeds compared to a rigid system. Comparedto classical exoskeletons, the system is lighter andless restrictive to the wearers motion. In addition,it can be put on and taken off very rapidly or sim-ply disengaged from the wearer while assistance isnot required. While having these advantages, thesystem is limited in the amount of assistance thatcan be applied to the wearer and uses a single motorthat only assists unidirectionally (in that it assistship extension but not flexion). However, we believethis will be sufficient for improving walking econ-omy.

For the current control strategy, we assumed thatan appropriate torque profile to apply to the hipwas a scaled version of the biological joint moment,but further work is required to validate this. Forassistive applications such as hiking or load car-riage, it may be beneficial to provide the wearerwith as much power as possible. For rehabilitation,individuals may require moment profiles specific totheir particular injury. If a scaled biological jointmoment is indeed best, increasing the speed andbandwidth of the actuation system would be use-ful to duplicate this moment more precisely, sincethe device currently provides a smoothed version ofthe biological moment. At force levels higher thanthose used currently, matching the biological mo-ment may be especially important. Increasing thebandwidth of the actuators could also permit usingforce-control instead of a position-control scheme.Improving the control scheme will be a major thrustof future work, so the system can adapt on a per-step basis to changes in gait. Adding additionalsensors such as gyroscopes to estimate the hip an-gle over time may be useful. Further testing of thesystem is also necessary to determine its effect ongait, both kinematically and metabolically.

In future work, the system will be made signifi-cantly lighter and smaller, so that it can be pack-aged compactly to fit onto a hiking backpack or or-thosis supporting the trunk. Further optimizationwill also be performed to select the vertical height ofthe actuators with respect to the hip joint to max-

imize the overall efficiency. The current design al-lows a person to sit down while wearing the device,but positioning the actuators lower may improvethe stiffness and increase the moments transmittedto the joints. Alternately, for some applications alower-profile unit that could be worn under cloth-ing may be beneficial. In all of these situations,the ability to provide moments at the hip througha compliant, non-restrictive mechanism should bevery useful.

7. Acknowledgements

This material is based upon work supported bythe Defense Advanced Research Projects Agency(DARPA), Warrior Web Program (Contract No.W911QX-12-C-0084). The views and conclusionscontained in this document are those of the au-thors and should not be interpreted as representingthe official policies, either expressly or implied, ofDARPA or the U.S. Government.

This work was also partially funded by the WyssInstitute for Biologically Inspired Engineering atHarvard University.

References

[1] A. Chu, H. Kazerooni, A. Zoss, On the biomimetic de-sign of the Berkeley lower extremity exoskeleton (bleex),in: Robotics and Automation, 2005. ICRA 2005. Pro-ceedings of the 2005 IEEE International Conference on,IEEE, 2005, pp. 4345–4352.

[2] H. Kazerooni, R. Steger, The Berkeley Lower Extrem-ity Exoskeleton, Journal of Dynamic Systems, Measure-ment, and Control 128 (2006) 14.

[3] C. Walsh, K. Endo, H. Herr, A quasi-passive leg ex-oskeleton for load-carrying augmentation, InternationalJournal of Humanoid Robotics 4 (03) (2007) 487–506.

[4] E. Garcia, J. M. Sater, J. Main, Exoskeletons for humanperformance augmentation (EHPA): A program sum-mary, Journal-Robotics Society Of Japan 20 (8) (2002)44–48.

[5] A. Esquenazi, M. Talaty, A. Packel, M. Saulino,The ReWalk powered exoskeleton to restore ambula-tory function to individuals with thoracic-level motor-complete spinal cord injury, American Journal of Phys-ical Medicine & Rehabilitation 91 (11) (2012) 911–921.

[6] P. D. Neuhaus, J. H. Noorden, T. J. Craig, T. Tor-res, J. Kirschbaum, J. E. Pratt, Design and evaluationof Mina: A robotic orthosis for paraplegics, in: Reha-bilitation Robotics (ICORR), 2011 IEEE InternationalConference on, IEEE, 2011, pp. 1–8.

[7] H. Yano, S. Kaneko, K. Nakazawa, S. Yamamoto,A. Bettoh, A new concept of dynamic orthosis for para-plegia: the weight bearing control (wbc) orthosis, Pros-thetics and orthotics international 21 (3) (1997) 222–228.

12

[8] B. J. Ruthenberg, N. A. Wasylewski, J. E. Beard,An experimental device for investigating the force andpower requirements of a powered gait orthosis, Journalof rehabilitation research and development 34 (1997)203–214.

[9] Y. Ohta, H. Yano, R. Suzuki, M. Yoshida,N. Kawashima, K. Nakazawa, A two-degree-of-freedommotor-powered gait orthosis for spinal cord injury pa-tients, Proceedings of the Institution of Mechanical En-gineers, Part H: Journal of Engineering in Medicine221 (6) (2007) 629–639.

[10] K. A. Strausser, T. A. Swift, A. B. Zoss, H. Kazerooni,B. C. Bennett, Mobile exoskeleton for spinal cord in-jury: Development and testing, ASME, 2011.

[11] H. Kawamoto, Y. Sankai, Power assist system HAL-3for gait disorder person, in: Computers helping peoplewith special needs, Springer, 2002, pp. 196–203.

[12] A. Dollar, H. Herr, Lower extremity exoskeletonsand active orthoses: Challenges and state-of-the-art,Robotics, IEEE Transactions on 24 (1) (2008) 144–158.

[13] H. A. Quintero, R. J. Farris, C. Hartigan, I. Clesson,M. Goldfarb, A powered lower limb orthosis for provid-ing legged mobility in paraplegic individuals, Topics inspinal cord injury rehabilitation 17 (1) (2011) 25–33.

[14] M. Arazpour, A. Chitsazan, S. W. Hutchins,F. T. Ghomshe, M. E. Mousavi, E. E. Takamjani,G. Aminian, M. Rahgozar, M. A. Bani, Design and sim-ulation of a new powered gait orthosis for paraplegicpatients, Prosthetics and Orthotics International 36 (1)(2012) 125–130.

[15] M. L. Audu, C. S. To, R. Kobetic, R. J. Triolo, Gaitevaluation of a novel hip constraint orthosis with im-plication for walking in paraplegia, Neural Systemsand Rehabilitation Engineering, IEEE Transactions on18 (6) (2010) 610–618.

[16] B. G. do Nascimento, C. B. S. Vimieiro, D. A. P.Nagem, M. Pinotti, Hip orthosis powered by pneumaticartifcial muscle: Voluntary activation in absence of my-oelectrical signal, Artifcial Organs 32(4) (2008) 317–322.

[17] C. Vimieiro, B. G. do Nascimento, D. A. P. Nagem,M. Pinotti, Development of a hip orthosis using pneu-matic artificial muscles, in: Technology Meets SurgeryInternational, Sao Paulo, Brazil, 2005.

[18] C. L. Lewis, D. P. Ferris, Invariant hip moment patternwhile walking with a robotic hip exoskeleton, Journalof Biomechanics 44 (2011) 789–793.

[19] J. A. Blaya, H. Herr, Adaptive control of a variable-impedance ankle-foot orthosis to assist drop-foot gait,Neural Systems and Rehabilitation Engineering, IEEETransactions on 12 (1) (2004) 24–31.

[20] G. S. Sawicki, D. P. Ferris, Powered ankle exoskele-tons reveal the metabolic cost of plantar flexor mechan-ical work during walking with longer steps at constantstep frequency, Journal of Experimental Biology 212 (1)(2009) 21–31.

[21] K. Yasuhara, K. Shimada, T. Koyama, T. Ido,K. Kikuchi, Y. Endo, Walking assist device with stridemanagement system, Honda R&D Technical Review21 (2) (2009) 57–66.

[22] Honda walking assist device with stride managementsystem, http://corporate.honda.com/innovation/walk-assist/, Accessed: December 20, 2013 (2013).

[23] M. Wehner, B. Quinlivan, P. M. Aubin, E. Martinez-Villalpando, M. Bauman, L. Stirling, K. Holt, R. Wood,

C. Walsh, Design and evaluation of a lightweight soft ex-osuit for gait assistance, in: Robotics and Automation,2013. Proceedings. ICRA’13. 2013 IEEE InternationalConference on, IEEE, 2013.

[24] A. T. Asbeck, R. Dyer, A. Larusson, C. J. Walsh,Biologically-inspired soft exosuit, in: RehabilitationRobotics (ICORR), 2013 IEEE International Confer-ence on, IEEE, 2013.

[25] T. Kawamura, K. Takanaka, T. Nakamura, H. Os-umi, Development of an orthosis for walking assistanceusing pneumatic artificial muscle: A quantitative as-sessment of the effect of assistance, in: RehabilitationRobotics (ICORR), 2013 IEEE International Confer-ence on, IEEE, 2013, pp. 1–6.

[26] T. Tanaka, Y. Satoh, S. Kaneko, Y. Suzuki,N. Sakamoto, S. Seki, Smart suit: Soft power suit withsemi-active assist mechanism-prototype for supportingwaist and knee joint, in: Control, Automation and Sys-tems, 2008. ICCAS 2008. International Conference on,IEEE, 2008, pp. 2002–2005.

[27] Y. Imamura, T. Tanaka, Y. Suzuki, K. Takizawa,M. Yamanaka, Motion-based design of elastic belts forpassive assistive device using musculoskeletal model, in:Robotics and Biomimetics (ROBIO), 2011 IEEE Inter-national Conference on, IEEE, 2011, pp. 1343–1348.

[28] H. Kobayashi, H. Nozaki, Development of anteversionposture support system, in: Robotics and Biomimetics,2008. ROBIO 2008. IEEE International Conference on,IEEE, 2009, pp. 682–686.

[29] J. G. Fontaine, Retracting device for automobile seatbelts, US Patent 3,199,804 (Aug 10 1965).

[30] E. E. Paananen, L. C. Csengery, R. J. Barry, Magnetictape drive, US Patent 3,764,087 (Oct 9 1973).

[31] H. Koselka, B. A. Wallach, D. Gollaher, Autonomousfloor mopping apparatus, US Patent 6,741,054 (May 252004).

[32] H. Zhu, C. Butts, M. Lamb, P. Blankenship, An imple-ment to install and retrieve surface drip irrigation lat-erals, Applied engineering in agriculture 20 (1) (2004)17–24.

[33] A. N. Lay, C. J. Hass, R. J. Gregor, The effects of slopedsurfaces on locomotion: a kinematic and kinetic analy-sis, Journal of Biomechanics 39 (9) (2006) 1621–1628.

[34] E. W. Weisstein, Archimedes’ spiral,http://mathworld.wolfram.com/ArchimedesSpiral.html,Accessed: Jan 2014 (2014).

13